Analyse the most important STATISTICAL characteristics of the pattern and choose a method of trading on it. - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Have you tried a confluent analysis? I.e. the function should not be price vs. time p = x(i), but two-dimensional f = z(i, p). Distance d is counted by two coordinates. And the other formulas are the same.
No, I haven't tried it, but it's interesting. In the end I decided that distortions along time axis (not only shrinking or stretching of patterns, which is linear distortion, but non-linear distortion too) should be taken into account according to principle our brain can recognize distorted representations of objects and people, even caricatures, that is by breaking patterns into components, their rotation, zooming etc. like in visual cortex. But little time has been spent on this. All the same, trading in the market, even on the most intricate mathematical models, will be 50/50.
No, I haven't tried it, but it's interesting. In the end I decided that distortions along the time axis (not only compression or stretching of patterns, which are linear distortions, but also non-linear distortions) should be taken into account by the principle that our brain can recognise distorted representations of objects and people, even caricatures, i.e. by breaking patterns into their components, their rotation, scaling, etc. as in the visual cortex. But little time has been spent on this. It would still be 50/50 to trade in the market even on the most intricate mathematical models.
Machine vision should handle this well, will do later
No, I haven't tried it, but it's interesting. In the end I decided that distortions along the time axis (not only compression or stretching of patterns, which are linear distortions, but also non-linear distortions) should be taken into account by the principle that our brain can recognise distorted representations of objects and people, even caricatures, i.e. by breaking patterns into their components, their rotation, scaling, etc. as in the visual cortex. But little time has been spent on this. All the same, trading in the market even with the most sophisticated mathematical models will be 50/50.
No, I haven't tried it, but it's interesting. In the end I decided that distortions along the time axis (not only compression or stretching of patterns, which are linear distortions, but non-linear distortions as well) should be considered according to the principle that our brain can recognize distorted images of objects and people, even caricatures, i.e. by breaking patterns into their components, their rotation, scaling and so on, like in the visual cortex. But little time has been spent on this. All the same, trading in the market, even on the most intricate mathematical models, will be 50/50.
Time should definitely be taken into account, yes. I, for example, do the following: I decrease the similarity of patterns as a function of time, linearly. I.e. if I estimate the similarity of the patterns from 0 (they are not similar at all) to 1 (they are completely similar), then in addition to the estimate I take away a constant multiplied by the number of bars between the patterns. I don't know what kind of distortions take place there, but the farther the patterns are from each other the more they lose their "similarity", 100% guarantee.
Similarity of patterns is a very exacting estimation, you cannot just plug the first formula from the internet, you have to check it by yourself. How to check the formula is also complicated and unclear, but some formulas will fail on fronttest, others will not :)
About pattern similarity - interesting example of how comp. vision works :)
http://cs.stanford.edu/people/karpathy/convnetjs/demo/mnist.html
How is your project with quarterly forecasts going? - The thread hasn't been updated in a while, it seems.
I put the latest predictions there. The last one was two months ago. The next one will be at the end of April with the release of new GDP data. So far all predictions are close to reality. I have two models there, one more conservative than the other. According to the conservative model the next GDP growth will be less than the growth published for the last quarter. The other model predicts higher growth. We will know in 3 weeks which one is more accurate. My main goal is to avoid a recession and so far it is not visible in either model.
Haven't looked here, maybe something will turn up ....Keldysh Library http://library.keldysh.ru/preprint.asp?id=2016-7 Orlov
Thanks, let's have a read
Let's say we have a piece of a chart. We need to work out (on the history) the best way to open deals on it. Where to buy, where to sell, where to buy more, where to close, and so on. But we must consider that the patterns may be different, and we must find the most effective method of calculating the position opening places for any pattern, while minimizing the risks. There can be several deals in a pattern. There is one more important condition, the pattern may vary within a certain range, say 20%. That is, at the beginning we see one pattern and on the next bar it will change somewhat, though its basic characteristics remain the same (but we will always see the whole pattern and all its future changes). That is, we need to introduce some other error factor.
Do you have any idea how best to do this? Various probabilities and price levels can be calculated, how can this be done?
It's interesting that in one of the discussions you were a staunch opponent of classical technical analysis, stating that its use is ineffective. The automation of manual trading based on this analysis is not acceptable. Now you have solved the problem of creating an effective method of algorithmic recognition of price formations, which is nothing else but an attempt to automate "manual" technical analysis. Strange, why have you just recently, vehemently rejected this approach in algotrading? (Pardon the off-topic).
Interestingly, in one of the discussions, you were an ardent opponent of classical technical analysis, stating that its use is ineffective. Manual trading automation based on this analysis is not accepted. Now you have solved the problem of creating an effective method of algorithmic recognition of price formations, which is nothing else but an attempt to automate "manual" technical analysis. Strange, why have you just recently, vehemently rejected this approach in algotrading? (Pardon the off-topic).
There's no classical tehanalysis, there's a multifractal model of asset returns (yes, yes, there's one too, I didn't just make that up). this model can be classified as a statmodel. There are no specific fixed patterns, but the result is a prediction, which can be represented as a pattern, nothing more.
MMDA describes a generalised Brownian motion